Your phone buzzes at 2:47 a.m. with a downtime alert, and the next 15 minutes decide whether this becomes a short blip or a long, expensive outage. This article is a practical website outage triage checklist for those first 15 minutes. It covers what to verify, what to rule out, who to tell, and when to stop investigating and start fixing.
The checklist is for small teams, solo developers and on-call engineers who don’t have a dedicated incident commander. You get a repeatable order of operations, so you aren’t refreshing a browser tab and guessing.
Why Early Triage Shapes the Whole Incident
Most outage time isn’t spent fixing things. It’s spent working out what is broken. A misconfigured Nginx reload takes 30 seconds to revert, but finding it can take 40 minutes if the responder starts in the wrong place.
Structured triage cuts that discovery time. It also stops the two most expensive moves under pressure: restarting everything at random, and making several changes at once so nobody knows which one helped.
Minute 0-2: Confirm the Outage Is Real
An experienced SRE first checks whether the alert reflects reality. Look at the alert itself: what status code came back, from which monitoring location, and how many consecutive checks failed? A single timeout from one region means something very different from five failed checks in a row across three locations.
Next, check from outside your own network. Run curl -I https://yourdomain.com from a laptop on mobile data, or from a cloud VM. Office VPNs, corporate DNS resolvers and browser caches all hide or invent problems. If the monitor says down and curl returns a clean 200, you may be looking at a regional routing issue, a WAF rule blocking the monitoring IPs, or a flapping health check. All three are real problems, but they aren’t the same as a site-wide outage.
Minute 2-5: Measure the Blast Radius
Before touching anything, work out the scope. Is it the whole domain or one path? Is the homepage up while /checkout returns 502? Does the API respond while the frontend is blank? Partial failures are common, and they point to different layers than a total blackout.
Check what changed recently. Look at the deploy log, CI pipeline history, DNS registrar audit log and CDN configuration changes from the last 24 hours. A large share of incidents trace back to a change made within the previous few hours, often by the same team now paged to fix it.
Then check your providers’ status pages. AWS Health Dashboard, Cloudflare Status, Google Cloud Status and your payment provider’s status page take 60 seconds to scan. If your hosting provider has acknowledged an incident in your region, your triage changes completely – the job becomes mitigation and communication, not debugging.
Minute 5-10: Walk the Stack from the Outside In
Work from the user’s side toward the database. Each layer either passes or fails, and the first failure is usually where to dig. This order holds up across most stacks:
DNS – run dig yourdomain.com +short against 1.1.1.1 and 8.8.8.8 and compare the answers. An empty answer or an unexpected IP suggests an expired domain, a broken zone edit or a registrar problem.
TLS – run openssl s_client -connect yourdomain.com:443 to see the certificate dates. A Let’s Encrypt certificate with a 90-day lifetime and a silently failed renewal job is a classic cause.
CDN or proxy – Cloudflare 521 means the origin refused the connection, 522 means it timed out, and 524 means the origin accepted the connection but took longer than 100 seconds to respond.
Web server and application – check systemctl status nginx or php-fpm, df -h for full disks, and journalctl or application logs for the first error, not the latest one.
Database – look for MySQL “Too many connections”, lock waits or replication lag, and PHP-FPM “server reached pm.max_children” warnings that point to a database bottleneck.
If DNS looks wrong, the guide on diagnosing DNS failures covers the common patterns in more depth.
Minute 10-15: Communicate and Choose a Path
By minute 10, someone outside the technical team needs to know. A two-line internal message is enough: what is affected, when it started, and when the next update is due. Customer-facing wording takes more care. The guide on communicating downtime to customers covers tone and timing.
Then decide between rollback and fix-forward. If a deploy went out in the last few hours and the symptoms fit, roll back first and investigate later. If no change explains it, pick the most likely layer from your stack walk and commit to it for a fixed window, say 10 minutes, before you re-evaluate.
A Scenario Most Teams Will Recognize
A WooCommerce store gets a downtime alert at 9:12 on a Monday morning. The homepage loads from the CDN cache, so the owner’s browser test looks fine. The monitor still reports 504 on /cart. The stack walk shows healthy DNS and a valid certificate, but PHP-FPM logs show workers maxed out. The cause is a plugin update applied automatically at 3 a.m., which added an unindexed query that only hurt once Monday traffic arrived.
Rolling back the plugin restored the cart in four minutes. Without the triage order, the owner would have trusted the cached homepage and assumed the alert was false.
The Myth That Root Cause Comes First
A common belief is that responders should find the root cause before acting. In the first 15 minutes, that’s backwards. The goal is to restore service, and root cause belongs in the post-mortem. Rolling back a deploy you haven’t fully diagnosed is a legitimate move if it gets customers back in.
There is one exception. If you suspect a security incident, such as defacement, unexpected admin accounts or outbound traffic spikes, preserve logs and disk state before restarting or rolling back. Wiping the evidence can create a bigger problem than the outage itself, especially in regulated industries under GDPR or PCI DSS breach-reporting rules.
Common Triage Mistakes to Avoid
The first mistake is trusting a browser check. CDN caching, service workers and local DNS caches regularly show a healthy page while real users get errors. Always confirm from outside your network with a cache-bypassing request.
The second mistake is changing several things at once. Restarting the web server, clearing the cache and bumping PHP memory in the same minute might fix the site, but nobody will know which change worked, and the same outage comes back next month.
The third mistake is going silent. Responders with their heads down for 30 minutes leave sales and support guessing, which leads to duplicate escalations and contradictory messages to customers. Writing these steps into a documented incident response plan removes that ambiguity before the next page arrives.
Website Outage Triage FAQ
Should I restart the server immediately when the site goes down?
Only after a quick look at logs and disk space. A restart can temporarily hide a memory leak or a full disk, and it erases the process state you need to understand what happened.
How do I know if the problem is my host or my code?
Check your provider’s status page and the time of your last deploy. If static files or a plain health-check endpoint also fail, suspect infrastructure. If only dynamic pages fail, suspect the application or database.
What if I’m a solo developer with no team to notify?
Notify your clients or stakeholders directly with a short holding message. Even a “we’re investigating, next update in 20 minutes” email saves you from a stream of interrupting calls.
Summary: Practice the Order Before You Need It
In the first 15 minutes of an outage, your job is to confirm, scope, walk the stack, communicate, and choose a path. Teams that do this well aren’t smarter under pressure. They’ve rehearsed the order. Print this checklist, run through it once during a quiet week against a staging site, and the next 2:47 a.m. alert will feel like a procedure rather than a panic.
